Based on prior animal and computational models, we propose a double dissociation between the associative learning deficits observed in patients with medial temporal (hippocampal) damage versus patients with Parkinson's disease (basal ganglia dysfunction). Specifically, we expect that basal ganglia dysfunction may result in slowed learning, while individuals with hippocampal damage may learn at normal speed. However, when challenged with a transfer task where previously learned information is presented in novel recombinations, we expect that hippocampal damage will impair generalization but basal ganglia dysfunction will not. We tested this prediction in a group of healthy elderly with mild-to-moderate hippocampal atrophy, a group of patients with mild Parkinson's disease, and healthy controls, using an “acquired equivalence” associative learning task. As predicted, Parkinson's patients were slower on the initial learning but then transferred well, while the hippocampal atrophy group showed the opposite pattern: good initial learning with impaired transfer. To our knowledge, this is the first time that a single task has been used to demonstrate a double dissociation between the associative learning impairments caused by hippocampal versus basal ganglia damage/dysfunction. This finding has implications for understanding the distinct contributions of the medial temporal lobe and basal ganglia to learning and memory.